Apply SQL Edits to Workspace Node

Round-trip a user's edits on a rendered DDL/DML SQL document back to the cloud workspace node it came from. Parses the SQL's column list, diffs it against the node's current metadata.columns[], and updates the node via set_workspace_node (unless dryRun: true). Both DDL and DML inputs feed the same node metadata — a DDL edit and a DML edit on the same node update the same metadata.columns[].

Column-level diff (structural changes — joins, CTEs, WHERE/GROUP BY, custom expressions on the join itself — are NOT reverse-engineered; those live in the node YAML):

• Removed columns (in YAML, missing in SQL) → dropped

• Type changes (matching name, different dataType) → updated in place; existing sources / columnReferences / lineage metadata is preserved

• Renamed columns (DML only — same transform, new AS alias) → renamed in place, lineage preserved (column id stays stable so downstream refs keep working). Detection requires unambiguous transform pairing; ambiguous cases fall back to drop+add.

• Expression-changed columns (DML only — same name, edited transform) → sources[].transform is updated and source-mapping inference is re-run to refresh columnReferences if the new transform references different upstream columns.

• New columns → for DML inputs, source-mapping inference resolves the column's transform against upstream predecessors (alias-prefix match, all-pred name fallback, multi-source COALESCE/GREATEST). For DDL inputs, new columns are added with empty columnReferences[] (DDL doesn't carry transforms); fill them in via the cloud UI.

Block-level diff (DML only — single-sourceMapping nodes):

• WHERE / ON-condition edits → written to sourceMapping[0].join.joinCondition with table refs rewritten to {{ ref('LOC', 'NAME') }} form.

• Adding or removing a predecessor source via SQL is REJECTED — add/remove the predecessor in the cloud UI first, then re-render.

Tail-clause diff (DML only):

• GROUP BY presence toggle → flips config.groupByAll. Coalesce derives the actual GROUP BY column list from non-aggregate columns at render time.

• ORDER BY adds / changes / removes → config.orderby flag and config.orderbycolumn.items[] array of {sortColName, sortOrder}.

• LIMIT adds / changes / removes → appended to / stripped from the trailing position of sourceMapping[0].join.joinCondition. Coalesce has no native limit config field for typical SQL nodes; the apply path warns when groupByAll/orderby is set (the auto-rendered tail clauses would land AFTER LIMIT, producing invalid SQL).

INSERT header diff (DML only — INSERT-shaped inputs):

• Target identifier change (INSERT INTO "DB"."LOC"."NAME") → updates the cloud node's top-level database / locationName / name fields. Renames/relocates the node; downstream references are unaffected (they use the node id, not the name).

• INSERT column list out-of-sync with the SELECT projection → REJECTED. Coalesce auto-generates the INSERT list from metadata.columns[]; edit the SELECT (using AS aliases) to add/rename/reorder columns.

Args:

• workspaceID (string, required)

• nodeID (string, required)

• sql (string, required): Either a CREATE [OR REPLACE] TABLE statement (DDL, from coa_dry_run_create) OR a SELECT/INSERT statement (DML, from coa_dry_run_run). The shape is auto-detected.

• dryRun (boolean, optional): If true, return the diff without writing.

Returns: { applied, dryRun, inputKind, diff: { unchanged, typeChanged, renamed, expressionChanged, removed, added }, joinDiff: { kind: 'identical' | 'whereOrJoinEdit' | 'newSource' | 'removedSource' | 'unsupported', ... }, tailDiff: { groupBy: { kind: 'identical' | 'added' | 'removed', ... }, orderBy: { kind: 'identical' | 'added' | 'changed' | 'removed', ... } }, limitDiff: { kind: 'identical' | 'added' | 'changed' | 'removed' | 'unsupported', ... }, insertHeaderDiff: { kind: 'identical' | 'targetChanged' | 'columnListMismatch' | 'notApplicable', ... }, body?, warnings?, error? }